Anthropic blocks access to its newest AI models after U.S. government directive

Anthropic said on Friday that it would block all users from accessing its most advanced AI models after receiving a U.S. government instruction tied to national security concerns. The models had been released only on Tuesday, and the company said the order came from the U.S. government on Friday evening.

According to Anthropic, the government letter did not include specific details about the alleged security risk. The company said its understanding was that U.S. officials believed there was a way to bypass safeguards meant to prevent the model Fable 5 from being used to identify software vulnerabilities.

Anthropic stressed that it disagreed with the government’s conclusions, but said it had to comply and therefore could not limit the restriction only to foreign nationals. In a statement, the company said, “So far, the government has given us only verbal evidence of a narrow, possible, non-universal exploit... essentially involving a request that the model read specific source code and fix software bugs... We do not agree that the finding of a potential narrow exploit should be grounds for revoking a commercial model released to hundreds of millions of people.”

The company said it would provide more details within 24 hours.